methods of defect inspection and detection for copper cmp · class 31 med multiple to low spots...

Cabot Microelectronics - Slide 0

Methods Of Defect Inspection and Detection For Copper

CMPChristopher Hawes, Defect Metrology EngineerShumin Wang, Copper Development Manager

Maria Peterson, Global OEM ManagerCabot Microelectronics

Ray Campbell, Senior Applications EngineerKLA-Tencor

Presentation given at Semicon China, 3/18/04 and CMPUG 5/5/04


Outline

• Project goal– Defect reduction, a collaborative effort by

consumables and equipment companies• Motivation

– Industry trend for reduced defects as technology nodes shrink

• Application of defect inspection– Copper slurry– Barrier slurry

• Summary


Project Goal Equipment and Consumables companies will jointly develop methodology for reducing CMP induced defectivity.

Why is (defect) methodology development important?

0

200

400

600

800

1000

1200

1400

2001 2002 2003 2004 2005 2006 2007

CM

P M

etal

Def

ect

0

10

20

30

40

50

60

70

80

Critical D

efect Size (nm)

• As feature size is reduced, what is considered a tolerable defect becomes intolerable.

• Generally speaking, intolerable defects are greater than half the feature size. For a 0.13 µm line that would be any defect greater than .065 µm (160 nm) in size.

Source: ITRS Roadmap


• We understand that detection tools have to constantly improve to keep up with industry demands, but if there is not an understanding of how to properly utilize the tool, the data collected is misleading or incorrect.

Understanding how to detect, classify, and eliminate defects is extremely important to CMC and KLA-Tencor.


Terms Defined

• AIT (Advanced Inspection Tool) - Darkfield (laser based) pattern wafer inspection tool. Detects the light a defect scatters whenilluminated with a laser.

• ADC (Automatic Defect Classification) - A software / hardware system that uses a set of user defined images to automatically classify defects.

• RTC (Real Time Classification) - A rough classification of defects that occurs during the AIT inspection scan, real time.

• Defect Class (Manual Class) - A name and number assigned to a type of defect (Example: Skipping/Repeating scratch is a Class 27 defect)

• Defect Bin (ADC Bin or Fine Bin, RTC Bin or Rough Bin) - A group of defect classes used by the classifier (Example: Bin 25 consists of Class25 (stitching scratch), Class 26 (razor scratch), and Class 27 (skipping/repeating scratch)


AIT II Operation

Keyboard

Prealigner

RoboticArm

ChuckCh 0

Laser

IlluminatorOptics

Ch 1

Ch 2

APS

OpticalMicroscope

Locator

A B C

• Die-to-Die comparison done to determine if die contains defects

• Defect must be doubly detected to be considered real, singly on edge

A is compared to B(B compared to A in parallel) C is compared to B(B is compared to C in parallel)

• Double Darkfield Inspection maximizes signal-to-noise ratio

• Defects detected in Ch.1 and Ch.2 are combined for total defects.


Recipe GenerationThe recipe generation step is a series of sub-steps (AIT/ADC/RTC):

Standard AIT IIrecipe setup

Optimize recipeKLA-Tencor BKMand Application Based

Optimize recipe

Review Wafers ADC RTC

Review Wafers

BaselineMultiple Recipes For Each Slurry

• Optimization (Multiple recipes and ADC) + Baseline (Multiple recipes and ADC for each slurry) = ~400 reviews (250 defects per review) = 100,000 reviewed defects

Review Wafers

Review Wafers

ADC Review Wafers


• Criteria For New Recipes:– Defect total on control wafers should be reasonable and contain defects of interest.

The defect type pareto should be comparable for both recipes. Increased defect detection in fine line arrays, still detecting in large features. Detect new defect types / smaller defects (Recipe optimized on a defect in 0.35 µm line array).

• Increase Sensitivity Increase Defect Capture• To Keep Total Count Reasonable Decrease Inspection Area• Decide to keep same number of die and reduce the number of sub-die regions

10 µm Pitch

90% Dense / 5 µm Pitch

Bond Pads

50% Dense / 0.35 µm line width


Each Slurry Type Has Its Own Recipe - Until recipe was setup on a wafer polished in current generation slurry and then used on a wafer polished in an experimental slurry, not obvious that each slurry needs it’s own recipe to truly improve that slurry from the previous generation.

General recipe can be used to screen slurry If true reduction needed, then specific recipe needed for that slurry type

0

0.5

1

1.5

2

2.5

3

3.5

4

Wafer Polished With Gen 1 Slurry Wafer Polished With Experimental CuSlurry

Nor

mal

ized

Def

ect C

ount

s

Recipe Generated On A WaferPolished With Gen 1 Slurry

Recipe Generated On A WaferPolished With Experimental CuSlurry


ADC Overview

• AIT detects defects, then ADC classifies defects• ADC classifies defects based on a set of user defined

images• ADC “sees” defects by doing a die-to-die comparison,

AIT can detect defects much smaller than ADC can “see” (Redetection Error)

• The “goodness” of the ADC will depend on the defect types, number of bins, and the types of examples chosen


Accuracy and Purity1 Man 5 Man 12 Man 13 Man 21 Man 25 Man Purity

1 ADC 244 0 84 124 69 12 0.465 ADC 9 0 8 21 2 2 012 ADC 7 0 104 155 29 40 0.4613 ADC 12 0 10 60 34 75 0.3121 ADC 60 0 84 25 570 360 0.5225 ADC 30 1 5 139 172 768 0.69252 ADC 12 0 3 26 16 5254 ADC 0 0 0 4 2 0255 ADC 0 0 2 0 0 0Accuracy 0.65 0 0.35 0.11 0.64 0.61

Overall Accuracy: 0.521Overall Purity: 0.538

ConfusionBin1: Type14 (Overpolish Barrier / Cu underpolish)Bin5: Type14 (Overpolish Barrier / Cu underpolish)Bin12: Type14 (Overpolish Barrier / Cu underpolish)Bin13: Type27 (Skipping/Repeating Scratch)Bin21: Type27 (Skipping/Repeating Scratch)Bin25: Type17, Type28, and Type31 (General Corrosion, Short Scratch, and Multiple Spots)Bin252: Type14 (Overpolish Barrier / Cu underpolish)

Accuracy =Number of Correctly Classified DefectsNumber of Manually Classified Defects

Purity =Number of Correctly Classified Defects

Number of ADC Classified Defects

•Accuracy is a measure of the ADC “capture rate” of a certain defect type

•Purity is a measure of the ADC“reliability” of certain defect type

*Source: KLA-Tencor IMPACT ADC Best Practices Document

*


Experimental Cu Slurry Recipe Comparison Total Defects

Normalized Defect Totals For Current and New Recipe - Experimental Cu Slurry

0.00

0.50

1.00

1.50

2.00

2.50

0 2 4 6 8 10 12 14

Wafer Number

Nor

mal

ized

Def

ect T

otal

Current Recipe Normalized DefectTotal

New Recipe Normalized DefectTotal


Experimental Cu Slurry Recipe Comparison Defect Distribution

• ~ 3250 defects manually reviewed for each recipe (recipe set for very high sensitivity)• Increased sensitivity and increased number of defects in 0.35 µm array can account for

changes in distribution - increase in Class 14 (die-to-die polish variation).• “Class 254” defects are bond pad roughness.• Major defect type not scratching, much improved over Gen 1 copper slurries.

23>22>3 14>23>26Class 23 - Multiple Residue on CuClass 22 - Single Residue on CuClass 3 - Multiple Residue on Dielectric

Class 14 Die-to-Die Polish VariationClass 23 - Multiple Residue on CuClass 26 - Razor Scratch on Cu

Manual Review Distribution For Experimental Cu Slurry Current Recipe

Manual Review Distribution For Experimental Cu Slurry New Recipe


Experimental Cu Slurry Recipe Comparison Defect Distribution

New recipe captures more defects in 0.35 µm array.Less inspected area in new recipe, but fine line array makes up a larger fraction

of the total area.

Defect Capture Comparison For Fine Line Arrays

02468

101214161820

Current Recipe New Recipe

Recipe

Perc

enta

ge o

f Tot

al D

efec

ts


Experimental Cu Slurry Recipe Comparison ADC Comparison

• Bin 252 is a standard ADC Bin (Redetection Error).• Within Bin 252 the class shifts from 26 (razor) to 14 (die-to-die polish variation)

ADC Distribution For Experimental Cu Slurry Current Recipe and Classifier

ADC Distribution For Experimental Cu Slurry New Recipe and Classifier


Experimental Cu Slurry Recipe Comparison ADC Comparison

Correlation Curve for ADC Bin 13 and Manual Review Bin 13 for New Recipe (Development Recipe) and Current Classifier

R2 = -3.9035

00.5

11.5

22.5

33.5

44.5

0 2 4 6 8 10 12 14 16

ADC

Man

Correlation Curve for ADC Bin 13 and Manual Review Bin 13 for New Recipe (Development Recipe) and New Classifier (Slurry Specific)

R2 = 0.9452

05

101520253035404550

0 5 10 15 20 25 30 35 40 45 50

ADC

Man


R2 = 0.9828

05

101520253035404550

0 10 20 30 40 50 60

ADC

Man

Correlation Curve for ADC Bin 21 and Manual Review Bin 13 for New Recipe (Development Recipe) and New Classifier (Slurry Specific)

R2 = 0.9375

0

20

40

60

80

100

120

140

160

180

0 20 40 60 80 100 120 140 160 180

ADC

Man


R2 = -4.5125

02468

1012141618

0 20 40 60 80 100 120 140 160

ADC

Man

Correlation Curve for ADC Bin 25 and Manual Review Bin 25 for New Recipe (Development Recipe) and New ADC (Slurry Specific)

R2 = 0.4763

0

1

2

3

4

5

6

7

0 2 4 6 8 10 12 14

ADC

Man

• Improvement in bins that contain corrosion, some improvement in scratch.• Currently working to further improve scratch bin, not the predominant defect type


Experimental Cu Slurry Defect Type Examples Defect Class 14 (Die-To-Die Polish Variation)

Note slight variation in color from die-to-die

Note over-polish variation from die-to-die


Experimental Cu Slurry Defect Type Examples

Defect “Class 254” (Roughness)Now referred to as Class 31 (Multiple Spots on Cu)

Roughness (3-11 A) remove by Barrier CMP


Experimental Cu Slurry Target Defect Type

Defect Type Number Impact"Class 254"Roughness High Low

Class 14Die-to-Die Polish High Low

VariationClass 23 High Low

Multiple ResidueClass 26 High Low

Razor ScratchClass 31 MedMultiple to LowSpots High

Class 21 Med LowDark / Hard Spot

Class 22 Med LowResidueClass 28 Low

Short Scratch MedClass 27

Skip/Repeat Low HighScratchClass 17 Low Low?Corrosion

• Defect map overlay work has revealed that most defects rated as low impact because Barrierpolish will remove them.

• Class 27 low number, but high impact.

• Class 27 (Skipping/Repeating Scratch on Cu) is the target defect type.

• Feature size where defects occur needs tobe taken into account.

• Experimental Cu slurry has ~8x less skip/repeat scratch defects and ~2.5x less total scratches than Gen 1 slurry.


Correlation Between ADC Bin 25 and Manual Review Bin 25 for New Recipe and New Classifier

R2 = 0.9144

0

5

10

15

20

25

30

35

40

0 5 10 15 20 25 30 35 40 45 50

ADC

Man

ual

Experimental Barrier Slurry ADC Comparison New Recipe

Correlation Between ADC Bin 25 and Manual Bin 25 for New Recipe and Old Classifier

R2 = 0.7941

05

101520253035404550

0 10 20 30 40 50 60 70 80

ADC

Man

ual

Correlation Between ADC Bin 21 and Manual Review Bin 21 For New Recipe and Old Classifier

R2 = 0.7786

0

5

10

15

20

25

30

35

40

0 10 20 30 40 50 60

ADC

Man

ual

Correlation Between ADC Bin 21 and Manual Review Bin 21 for New Recipe and New Classifier

R2 = 0.7691

0

5

10

15

20

25

30

35

40

45

0 10 20 30 40 50 60 70

ADC

Man

ual


Experimental Barrier Slurry Target Defect Type

Defect Type Number ImpactClass 26 High High/Med?

Razor ScratchClass 28 Medium Med?

Short ScratchClass 21 Medium Med?

Dark / Hard SpotClass 31Multiple Medium Med?Spots

Class 27Skip/Repeat Medium High

ScratchClass 25 Medium Med?

Stitching ScratchClass 22 Low Med?ResidueClass 14

Die-to-Die Polish Low Low/Med ?VariationClass 12 Low HighFall-on

Class 17 Low HighCorrosion

• Many defects are classified as high impact after Barrier polish, because this is what the customer sees.

• Overlay of Cu and Barrier polish needs to be doneto identify which defects occur at which step, and toidentify which defects Barrier polish removes.

• Barrier polishing erases Razor and Stitching scratches,and sometimes leaves remains of Skipping/Repeating behind from Cu polish as Multiple Spots.

• Target defects should be Class 26 (Razor Scratch on Cu) and Class 27 (Skipping/Repeating Scratch on Cu).

• Development work in the last 12 months has shown a a >50% reduction in razor scratch defects.


Performance Monitoring• After baseline is complete, Klarity Defect will be used to monitor experiment results.• Klarity Defect also used to differentiate between defects generated after Cu polish and after Barrier polish.


Summary

• Consumable suppliers and metrology equipment suppliers are jointly developing a CMP defect reduction methodology by joining skills of slurry optimization and inspection method optimization.

• To truly optimize a slurry, a unique inspection recipe is required. Recipes (classifiers) have to be modified as slurry improves. SEM inspection essential to keep accuracy high.

• Total number of defects important, knowing the distribution of the defect types even more important.

• Many experimental Cu slurry defect types removed by barrier slurry polish and the major defect type is not scratching, a large improvement over first generation slurries.


Acknowledgements• Copper Development TeamsCabot Microelectronics• Paul FeeneyIntegration Manager, Cabot Microelectronics• Jose EstabilSenior Director Technology IBM/EUSA Business Units, KLA-Tencor• Kelley HalchuckSales Account Manager, KLA-Tencor• James ZimmermanField Application Engineer Defect Solutions Division, KLA-Tencor• Mulia MandiroApplications Engineer - ADC, KLA-Tencor

Semi and NCCAVS – for sponsoring these symposia

methods of defect inspection and detection for copper cmp · class 31 med multiple to low spots...

Documents